AI rally narrows to infrastructure

The AI trade is spreading beyond chips and into the pipes and wires that keep those chips running. (markets.businessinsider.com) On April 22, Axe Compute said it signed a 36-month contract worth about $260 million to deploy 2,304 Nvidia B300 graphics processors and high-speed storage at a single Tier 3 data center in the United States. The company said the site will start deployment in the third quarter of 2026 and has 4.8 megawatts of dedicated power on an N+1 redundant basis. (markets.businessinsider.com) Nvidia’s own product materials describe the B300 as part of its Blackwell Ultra line for “AI factory” systems, with the DGX B300 pitched as a data-center platform for large-scale training and inference. Nvidia says the DGX B300 delivers 1.5 times the dense FP4 performance of the prior DGX B200 and is built to fit modern data centers. (nvidia.com) Power is moving up the stack with the chips. Nvidia said in May 2025 that it was leading a shift to 800-volt direct-current data-center power for 1-megawatt information-technology racks and beyond, starting in 2027. (developer.nvidia.com) A report published on April 22 by Asiae said Nvidia had approached South Korean power-equipment makers about 800-volt direct-current infrastructure, as global technology companies push for larger and faster power delivery to AI facilities. The report said Nvidia was proposing 800-volt direct current as a new standard for next-generation data centers. (asiae.co.kr) That buildout is landing in a market where electricity, cooling and backup systems already account for a large share of data-center design. The International Energy Agency said servers use about 60% of electricity in modern data centers, while cooling ranges from about 7% in efficient hyperscale sites to more than 30% in less-efficient enterprise facilities. (iea.org) The same International Energy Agency report estimated data centers used about 415 terawatt-hours of electricity in 2024, equal to about 1.5% of global power consumption. That makes grid access, transformers, switchgear, batteries and cooling equipment part of the same spending cycle as the processors themselves. (iea.org) South Korea is already one of Nvidia’s larger infrastructure partners. Nvidia said on October 30, 2025 that it was working with the South Korean government and companies including Samsung Electronics, SK Group, Hyundai Motor Group and NAVER Cloud on more than 260,000 Nvidia graphics processors across sovereign clouds and AI factories. (nvidianews.nvidia.com) For investors, that leaves the AI trade tied to a narrower set of bottlenecks than chip demand alone: one U.S. site in Axe Compute’s case, and a short list of power and equipment suppliers in Nvidia’s 800-volt push. The next leg of the rally now depends not just on how many chips can ship, but on how much electricity and infrastructure can be lined up around them. (markets.businessinsider.com; developer.nvidia.com; iea.org)